Transaction Detection

The Minimum Line Count property sets a threshold for the smallest number of lines that a detected section (transaction) must contain in order to be included in the extraction results. This helps filter out sections that are too small to be valid transactions, which may result from noise, misclassification, or incomplete data.

• If set to 0, all detected sections will be returned, regardless of their line count.
• If set to a positive integer (e.g., 1), only sections with at least that many lines will be included.

Usage Guidance

• Increase this value to exclude very short or empty sections that are unlikely to represent valid transactions.
• Use a value of 1 or higher for most transaction-based documents to avoid returning empty or spurious sections.
• For documents where valid transactions may be very short, set this value accordingly.

Example

If Minimum Line Count is set to 3, only sections containing three or more lines will be returned by the extractor.

This property is especially useful for improving the quality of extraction in documents with variable or noisy layouts.

The Minimum Confidence property sets a threshold for the quality of detected sections (transactions) based on the extractor's internal scoring. Only sections whose confidence score meets or exceeds this value will be included in the extraction results.

• Confidence is calculated based on how well the detected section matches expected periodicity, binding field alignment, and other extraction criteria.
• The value is expressed as a percentage (e.g., 0.75 for 75% confidence).
• Sections with a confidence below this threshold are discarded.

Usage Guidance

• Increase this value to require higher certainty in detected sections, reducing the risk of false positives.
• Lower this value if valid transactions are being missed due to low confidence scores, but review results for possible noise.
• The default value is typically 0.75 (75%), which balances accuracy and inclusiveness for most document types.

Example

If Minimum Confidence is set to 0.90, only sections with a confidence score of 90% or higher will be returned.

Adjust this property to fine-tune the sensitivity of transaction detection for your specific document set.

The Line Periodicity Detector is used to analyze a sequence of lines within a document and identify repeating patterns based on configurable similarity and alignment criteria. This is especially useful for extracting structured data from documents where information is organized in rows or sections that repeat.

How It Works

The detector compares lines of text to find groups that are sufficiently similar, as determined by the 'Similarity Threshold' and 'Minimum Alignment' properties. When a repeating pattern is detected, the system can treat each occurrence as a distinct section or transaction, enabling downstream extraction and processing.

Configuration Guidance

• Adjust the 'Similarity Threshold' to control how closely lines must match to be considered part of the same repeating pattern.
• Use 'Minimum Alignment' and 'Minimum Segment Length' to fine-tune the sensitivity of pattern detection, especially in documents with variable content or formatting.
• Optionally, assign a 'Background Extractor' to filter out known background elements that should not influence periodicity analysis.
• The 'Page Depth' property limits the number of pages analyzed, which can improve performance on large documents.

Only the first 5 pages of the document will be analyzed for periodicity.

Notes

• The Line Periodicity Detector is typically used as part of a section extraction or transaction detection workflow.
• Proper configuration is essential for accurate detection; test with representative samples to ensure optimal results.
• For more information, see the documentation for Section Extract Methods and Value Extractors.

Can be one of the following types:

The Transaction Layout Detection class enables Grooper to identify the start and end of transactions or sections within a document by considering visual presentation elements. This includes horizontal and vertical lines, boxes, and whitespace areas that often separate or group related content.

How It Works

When extracting transactions, this class uses the configured 'Minimum Width' and 'Maximum Distance' properties to determine which lines of text should be grouped together as part of the same transaction. By leveraging layout cues, it can more accurately segment complex documents where text alone is insufficient for reliable separation.

Configuration Guidance

• Set 'Minimum Width' to exclude narrow sections that are unlikely to represent valid transactions.
• Adjust 'Maximum Distance' to control how far apart lines of text can be and still be considered part of the same transaction.
• Both properties accept values in logical units (such as inches or percentages), allowing for flexible configuration across different document types and layouts.

Usage Example

To configure Transaction Layout Detection for a document with wide transaction rows and moderate spacing:

This configuration will only return sections that are at least 75% of the page width and will group lines that are within half an inch of each other.

Notes

• Transaction Layout Detection is typically used in conjunction with other section or transaction extractors for best results.
• Proper configuration is essential for accurate detection; test with representative samples to ensure optimal results.
• For more information, see the documentation for Section Extract Methods and Transaction Detection.

The Binding Fields property specifies which Data Fields are used as unique identifiers (binding values) when splitting a document into transactions using the Transaction Detection extractor. These fields play a critical role in ensuring that each detected Section Instance contains exactly one unique value for each binding field, which is essential for accurate and reliable sectioning of repeating document content.

How Binding Fields Are Used

During extraction, the Transaction Detection extractor analyzes the document for repeating groups of lines or label patterns. When multiple possible groupings are detected, the extractor evaluates each possibility by checking how well the binding field values are distributed:

• The preferred grouping is the one where each Section Instance contains exactly one unique value for each binding field.
• If a binding field value appears in more than one section, or is missing from a section, the grouping is penalized or rejected.
• This approach helps prevent over- or under-segmentation, especially in documents where transactions are similar in structure but may vary in length or content.

Configuration Guidance

• Select Data Fields that uniquely identify each transaction or repeating group in your document. Common examples include claim numbers, invoice numbers, account numbers, or other fields that are guaranteed to be present and unique within each transaction.
• You can select one or more fields. If multiple fields are chosen, the extractor will require that each section contains a unique combination of all binding field values.
• Avoid selecting fields that are optional, frequently missing, or not unique within the transaction, as this can lead to incorrect sectioning.

Example

For an Explanation of Benefits (EOB) document where each claim is identified by a "Claim Number" field:

Binding Fields: Claim Number

For a transaction that is uniquely identified by both an "Account Number" and a "Transaction Date":

Binding Fields: Account Number, Transaction Date

Impact on Extraction

• The accuracy of transaction detection is highly dependent on the correct configuration of binding fields.
• If no binding fields are specified, or if the selected fields do not uniquely identify each transaction, the extractor may produce incorrect or ambiguous results.
• When troubleshooting sectioning issues, review the binding fields to ensure they are present, unique, and reliably extracted in your documents.

For more information on configuring transaction detection, see the documentation for Transaction Detection, Data Section, and Data Field.

The Detect Page Spanning property determines whether the Transaction Detection extractor will attempt to merge section instances that represent the same transaction but are split across consecutive pages. This is especially important for documents where a single transaction may continue onto a new page, and the identifying binding values are repeated at the top of each page.

How It Works

When enabled, the extractor examines adjacent section instances at page boundaries. If the last transaction on one page and the first transaction on the following page have identical values for all configured binding fields, they are considered to be parts of the same transaction and are merged into a single Section Instance.

• If Detect Page Spanning is disabled, each occurrence of the binding values on a new page is treated as a separate transaction, even if the values are identical.
• If enabled, the extractor will combine these into a single section, provided all binding field values match.

This logic is useful for documents such as statements, EOBs, or reports where transactions may be split by page breaks but are clearly identified by repeated binding values.

Configuration Guidance

• Enable this property if your documents frequently contain transactions that span multiple pages and use repeated binding values for identification.
• Leave this property disabled if each page should always be treated as a separate transaction, or if repeated binding values do not indicate continuation.

Example

Suppose a claim in an EOB document starts on page 1 and continues onto page 2, with the "Claim Number" field repeated at the top of both pages. If Detect Page Spanning is enabled, these two sections will be merged into a single Section Instance for that claim:

Impact on Extraction

• Enabling this property helps ensure that transactions split by page breaks are not incorrectly treated as separate items.
• If disabled, you may see duplicate or fragmented data for transactions that span multiple pages.